Research On Key Technologies Of Social Recommendation System Based On Trust Prediction In Big Data Environment

In the era of big data,the contradiction between explosive growth of network information and sparsity of rating information hinder the development of traditional recommendation system.At the same time,malicious users and attacks that spread false information and illegally access personal privacy occur frequently on the network,which makes the construction and evaluation of trust relationship among network users more urgent.In recent years,academic and industrial circles have conducted extensive studies on social recommendation based on trust,but there are still some challenges and difficulties in the big data environment:lack of direct trust relationship between users and inaccurate trust prediction;the problem of single source of trust prediction model evaluation;the poor coupling between recommendation system and trust prediction model.This paper focuses on the issues related to trust prediction and social recommendation in the big data environment.Our research is carried out from three aspects:how to design a lightweight and efficient social network trust prediction method,how to integrate recommendation system and knowledge graph to assist in trust prediction,and how to combine social network trust relationship with recommendation system rating relationship to improve recommendation effect.Corresponding new methods and models are proposed.The main work of this paper are as follows.(1)Aiming at the problems of sparsity of direct trust relationship and inaccurate trust prediction in big data environment,a graph-based lightweight trust prediction method for social networks was proposed.Firstly,a locally weighted centrality measure is constructed based on the local network topology and trust relationship,and an adaptive breadth-first trust path search algorithm is proposed.Then,by exploring the decay trend and degree of trust propagating in a single path,the corresponding decay calculation method is proposed.Finally,a dynamic weighted aggregation strategy based on local weighted centrality measure is used to calculate the comprehensive trust values of multiple trust paths.On the basis of improving the accuracy of indirect trust relationship evaluation,the proposed trust prediction model improves the inefficiency of existing methods,subjective allocation of trust factors,and inconsistent opinions of trust neighbors.Experimental verification and parameter analysis on the real social dataset Advogato show that the proposed LWCTrust model outperforms some classical trust prediction models in both efficiency and accuracy.(2)To solve the problem of single source of trust prediction model evaluation,a trust prediction scheme based on recommendation system and knowledge graph as the additional information is proposed.On the one hand,This paper proposes a trust prediction method in social networks which takes the rating information of recommendation system as additional information.Through the matrix factorization method based on social and interest regularization,fully mining the neighbor trust relationship in the social network and the rating preference relationship in the recommendation system.By adding regularization constraints of social and interest neighbors,the accuracy of trust matrix factorization can be improved.On the other hand,a trust prediction method based on triplet information of knowledge graph as additional information is explored.By applying the knowledge representation learning method in knowledge graph to trust prediction,users and their relationships in social network are represented as vectors,and the corresponding distance index and ranking are calculated according to triangle rule to predict the trust value of the relationship between two unknown users.Experimental results on Epinions and Ciao real datasets show that the proposed model has better trust prediction accuracy than the traditional trust prediction method and the MF-based trust prediction method.(3)Aiming at the problem of poor coupling between recommendation system and trust prediction model,a social recommendation model based on trust prediction is proposed.Firstly,a comprehensive weighted centrality metric based on local and global context is proposed to identify authoritative and reliable users to resist malicious attacks and improve model efficiency.Secondly,a trust prediction algorithm based on trust propagation and aggregation strategy is proposed to complete the social trust graph.Finally,in a social recommendation matrix factorization model,the implicit influence of rating and trust as well as the comprehensive influence of truster and trustee are considered.We combine the rating matrix with direct trust matrix and indirect trust matrix,and propose a comprehensive social recommendation model based on matrix factorization to solve the sparsity problem of rating and direct trust information and the user cold start problem.Experiments on real datasets FilmTrust,Epinions and Ciao show that our proposed model has better rating prediction accuracy than the other matrix factorization based social recommendation model in both the "all users" view and the "cold start users" view.